Encord and the Scarcity of Physical Training Data
The scarcity of real-world physical training data is the next real constraint on humanoid and warehouse robotics, according to Encord, a company that builds data tooling used to train AI models. This mirrors what happened in the early days of self-driving cars, where companies like Waymo had to collect and annotate vast amounts of data to train their models. Encord’s work with Zander Labs, a German neuroscience startup, is a trial run to build an initial brain wave-tagged data set to improve robotics model performance.
Encord’s head of robot learning, Vineeth Velmurugan, says the company’s customers, including leading robotics firms, are struggling to find the data they need to train their models. “The data simply does not exist,” Velmurugan said. This is a major challenge for companies building robot brains, as they require vast amounts of high-quality data to train their models.
Encord is manufacturing data through various methods, including egocentric video collection and remote robot operation. The company is also experimenting with new modalities, such as brain waves and muscle sensors, to create more useful data sets. This is a significant shift in the industry, as companies are now turning to data generation as a business, rather than just a research problem.
Encord’s Decision Logic and Mechanics
Encord’s decision to focus on data generation is driven by the company’s incentive to create a scalable and profitable business model. By manufacturing data, Encord can provide its customers with the high-quality data they need to train their models, while also generating revenue through data licensing and annotation services. This is a strategic move, as Encord can now control the supply of data and dictate the terms of the market.
From a technical perspective, Encord’s data generation methods involve collecting and annotating data from various sources, including cameras, sensors, and human operators. The company is using machine learning algorithms to process and analyze the data, and to identify patterns and trends that can be used to improve robotics model performance. This is a complex task, requiring significant expertise in computer vision, machine learning, and data annotation.
Encord’s use of brain waves and muscle sensors is a notable innovation, as it allows the company to capture more nuanced and detailed data about human behavior and decision-making. This data can be used to train more sophisticated robotics models that can mimic human-like behavior and decision-making.
Winners, Losers, and Disrupted Parties
The winners in this scenario are companies like Encord, which are well-positioned to capitalize on the growing demand for high-quality data. Robotics companies that can access and utilize this data will also benefit, as they will be able to train more sophisticated models and improve their products and services. The losers are companies that are unable to access or generate high-quality data, as they will struggle to compete in the market.
Adjacent markets, such as computer vision and machine learning, will also be impacted by this development. Companies that specialize in these areas will need to adapt to the changing landscape and find ways to integrate with the new data generation models. Job categories, such as data annotation and robotics engineering, will also be affected, as new skills and expertise will be required to work with the new data generation technologies.
The downstream effect of this development will be significant, as it will enable the creation of more sophisticated and autonomous robotics systems. This will have major implications for industries such as manufacturing, logistics, and healthcare, where robotics systems can be used to improve efficiency, productivity, and safety.
The Skeptical Case
A skeptical view of this development is that it may be overhyped, and that the challenges of generating high-quality data are underestimated. This mirrors what happened with the rise of LLMs, where the promise of revolutionary progress was not fully realized. The assumption that brain waves and muscle sensors will provide a significant improvement in data quality may not hold, and the cost and complexity of generating this data may be prohibitively high.
A historical analogue to this scenario is the rise and fall of the dot-com bubble, where companies overpromised and underdelivered on their technological capabilities. A similar outcome is possible in this scenario, where companies may overpromise and underdeliver on the capabilities of their data generation technologies.
The Signal to Watch Next
The next verifiable event to watch is Encord’s announcement of its initial brain wave-tagged data set and its evaluation of its effectiveness in improving robotics model performance. This will provide a concrete indication of the success of Encord’s data generation strategy and its potential to disrupt the market.
Another signal to watch is the adoption of Encord’s data generation technologies by leading robotics companies. If these companies begin to integrate Encord’s technologies into their products and services, it will be a strong indication of the success of Encord’s strategy and its potential to disrupt the market.
What’s your take on this? Drop your perspective in the comments below.
By Alex Mercer, Senior Tech Analyst at TrendFlashy
Ready to launch your own asset?
Check out our guide on Building a Profitable Online Business.
